Tick-borne diseases and tick bites are an increasing concern in the Netherlands. Among all the environmental factors, vegetation plays an important role in tick life circle by providing shelter, water, a place for seeking hosts, etc. Knowledge about relationships between vegetation structure patterns and tick occurrence patterns can help mapping potential tick distribution and contribute to improve local risk management of ticks by the public health service in Amsterdam. It is also required to have vegetation structure map for monitoring and risk management of ticks in Amsterdam.
In this study, firstly grass, shrubs, and trees three vegetation structure categories were defined. Height and height spreads were used as criteria for classification of point cloud data created by Semi Global Matching of aerial photographs. The topographic dataset KBKA10 provided locations of vegetation for filtering the point cloud data. AHN2 terrain grid offered terrain height values for extracting vegetation area objects height. The aerial image of Amsterdam was used for providing ground information and selecting training locations of the three classes.
Then supervised classification with k-Nearest Neighbour and Classification And Regression Tree (CART) analysis were applied for mapping vegetation structure classes. Accuracy was assessed by computing confusion matrices. Thirdly, logistic regression was conducted with tick data, vegetation structure classes from the CART result maps, distance to water area and distance to built-up area to assess relationships between these factors and tick occurrence patterns in Amsterdam. The model was finally used for spatial prediction of tick occurrenceThe results show CART produced most accurate vegetation structure maps. The overall classification accuracy of CART result was 86%. The grass class had a high accuracy 97% and high reliability 93%. The tree class had a high reliability 99%. The overall classification accuracy of the k-Nearest Neighbour result was 77%. The grass class had 99% accuracy. The tree class had a reliability approaching 100%. Distance to water and distance to built-up area were found not to be significantly related to tick absence and presence in Amsterdam. The significant predictor vegetation structure classes influence tick occurrence in Amsterdam to a small extent. I found a weak association between vegetation structure classes and tick presence and absence in Amsterdam. The predicted tick presences appear at sites where shrubs or trees grow.
Keywords: Vegetation structure map; Supervised classification; k-Nearest Neighbour; Classification And Regression Tree; Tick occurrence; Logistic regression; Amsterdam